AIoT Enhanced Night Vision Object Detection for Vehicle Safety using Transfer Learning on Raspberry Pi5

We are now in the twenty-first century, but car safety issues persist, ensuring vehicle safety, especially in low-light conditions, is vital for a modern transportation system. Among these challenges, one of the major problem is detecting objects clearly and accurately at night, which significantly affects driver's response time and overall safety. The primary idea behind this Artificial Intelligence of Things (AIoT) based object detection project is to utilize the computational capabilities of the Raspberry Pi 5 as well as the YOLO (You Only Look Once) v5 through v8 advanced object detection algorithms. The goal is to achieve higher performance and efficiency of the vehicular object detection system without increasing computational resources. The experiment is implemented with the help of an infrared camera to grab low-light images, which are further pre-processed and then input into the YOLO models that were previously trained using transferred learning. The software part is automatic. By using the system time, the system turns on the night mode when it's dark and switches between traffic and highway modes when necessary. To this end, the technology is scalable with fine-grain identification and positioning (DLA-Id). This setup involves the use of the latest version of a Raspberry Pi 5, an infrared camera, a buzzer, a breadboard, and an LCD, all of which have been integrated and form a compact and efficient prototype. The results indicate that the device facilitates real-time object detection and distance calculation an accuracy within ±20 cm, along with identifying lane kinds and directions under all meteorological and driving situations, demonstrating its reliability under all meteorological and driving conditions.